A Simple Approach for Sample Size and Power Calculations for Clustered Count Data in Matched Cohort Studies

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Dexiang Gao*+ and Gary Grunwald and Stanley Xu
• Companies: University of Colorado at Denver and University of Colorado at Denver and Kaiser Permanente Colorado
• Keywords: Matched cohort design ; Clustered count data ; Random cluster effects Poisson model ; Sample size ; Statistical power

Abstract

In matched cohort studies treated and untreated individuals are matched on certain characteristics to form clusters (strata) to reduce potential confounding effects. In this study design clustered count data are often the outcomes being used to estimate the treatment effect. Random cluster effects Poisson models (RCP) are frequently used for analyzing the clustered count data. However, sample size and power calculation can be challenging because the within cluster correlation needs to be considered but generally is not available in planning phase of a study. In this paper we compare the treatment effect estimate and its variance from RCP and those from other models. We then propose a simple approach for calculating statistical power and sample size for clustered count data in matched cohort studies with a constant matching ratio. Preliminary results from simulations showed that the power and sample size calculations are accurate when the random cluster effects are from either normal or gamma distributions. We also evaluated the simple approach of power and sample size calculations when the matching ratio is varying across strata.